OpenAI models have boosted AMD chip design performance, power, and area by up to 14 percent. This collaboration marks a significant breakthrough, accelerating development timelines by achieving a year of progress in under five months.

OpenAI AI Models Boost AMD Chip Design Efficiency by 14 Percent Gains

New Delhi [India], November 19 (ANI): OpenAI said its artificial intelligence models have delivered up to 14 per cent improvements in chip design performance, power and area in collaboration with AMD, with some of the changes already incorporated into designs sent for manufacturing. The company said the improvements were achieved on AMD's flagship graphics processing unit (GPU) intellectual property, relative to heavily optimised baselines. It described the progress as “roughly a year of progress achieved in less than 5 months”, calling the results unprecedented at the frontier of chip design. AMD has already included some of the model-generated changes in a recent design sent for manufacturing, while others have been added to its production design files for future versions, OpenAI said. The models also helped improve key blocks of AMD's upcoming next-generation flagship central processing unit (CPU), with the company saying AMD had verified single-digit percentage gains in performance, power and area. OpenAI said the models were able to rewrite the code describing chip circuits, known as SystemVerilog RTL, while using Synopsys tools to verify that the changes preserved the original design's behaviour and assess their impact on power, performance and area. The company also reported improvements in its first custom inference chip, Jalapeño. Model-proposed changes reduced standard-cell area by around 13 per cent in its vector unit and 10 per cent in its matrix unit. These improvements were validated during the chip's silicon bring-up, it said. Separately, OpenAI evaluated successive generations of its models on two open-source RISC-V processor designs, PicoRV32 and Ibex. GPT-6.2 Astra recorded the highest average scores on both designs under comparable testing budgets, outperforming earlier models in optimising existing processor designs and developing new ones from written specifications. The evaluation covered 9,881 scored runs across the two processor designs and different testing budgets. OpenAI further said it is partnering with chip design software maker Synopsys to train a new model, GPT-Synopsys, to use electronic design automation tools in the way expert chip designers do. The collaboration is aimed at expanding the range of chip design tasks that AI models can handle and improving their evaluation against trusted engineering tools. (ANI)